Machine learning approach to predict protein phosphorylation sites by incorporating evolutionary information

BMC Bioinformatics

Table 4 Three fold cross validation performance of the prediction system using ratio 1:1.

Residue	W	Ac(%)	Sn(%)	Sp(%)	Mcc	FPR
	7	73.92	70.81	76.69	0.48	0.23
	9	74.19	70.02	77.92	0.48	0.22
S	11	74.41	70.16	78.20	0.49	0.22
	13	74.75	71.63	77.54	0.49	0.22
	15	74.83	73.12	76.35	0.50	0.24
	7	69.54	65.27	73.69	0.39	0.26
	9	69.71	66.49	72.84	0.40	0.27
T	11	70.18	67.52	72.76	0.41	0.27
	13	70.02	66.85	73.11	0.41	0.27
	15	70.22	68.31	72.07	0.41	0.28
	7	67.67	68.51	66.86	0.36	0.33
	9	68.74	67.80	69.66	0.38	0.30
Y	11	68.53	68.24	68.81	0.38	0.31
	13	68.45	68.07	68.81	0.37	0.31
	15	69.44	70.37	68.55	0.39	0.31

Using this ratio greatly balances the two accuracy parameters - Sensitivity and Specificity along with the Mathews correlation coefficient (Mcc). All the accuracy parameters show better performance if window size increases, i.e., more features are added to each of the training instances for the SVMs.

ISSN: 1471-2105